You are right about expected value (average), however you also have to consider distribution.
1k12 means every value is equally possible to roll. You get 1 as often as 3, 8 or 12. Exactly 8,33%.
2k6 is normal distribution (Gauss or bell curve). The most possible option to roll is 7 - 16,7%. It's 2 times more probable when compared to rolling 1k12. And there is only 2,78% to roll 2 or 12. Nearly 3 times less possible.
Rolling 2k6 is "safer" option to roll. There is almost 45% chance to roll 6, 7 or 8.
Of course my math is only accurate if game engine is actually rolling two (or more) dices. If implementation of 2k6 is done by rolling one k6 (as double precision number) and just multiplying the value by 2 then... there is almost no difference than rolling 1k12. Only this +0,5 to expected value as SecSea mentioned, however you will hardly notice that.
